QPF sensitivity to Runge-Kutta and Leapfrog core

1. Axel Seifertwith Jochen Förstner, Günther Zängl, Michael Baldauf Deutscher Wetterdienst, Offenbach Deutscher Wetterdienst GB Forschung und Entwicklung QPF sensitivity to Runge-Kutta and Leapfrog core

2. QPF with COSMO-EU: LF vs RK core Known problems and biases: • Operational COSMO-EU shows a strong overestimation of stratiform (grid-scale) precipitation during winter. • COSMO-EU with most recent version of the Runge-Kutta numerics reduces the wintertime bias, but the reduction seems to be too strong leading to an underestimation of precipitation. Outline of this talk: • QPF verification comparing Leapfrog and Runge-Kutta in COSMO-EU • Microphysics experiments: re-tuning necessary for RK-core? • Conclusions

3. Accumulated precip April 2009: REGNIE LF COSMO-EU RK-core suite 75 mm 101 mm (+35%) 66 mm (-12 %) • Operational COSMO-EU with LF shows strong positive bias. • COSMO-EU with RK-core reduces the precip amount, but leading to a negative bias. Spatial distribution is too smooth, orographic enhancement is underestimated.

4. Accumulated precip August 2009: REGNIE LF COSMO-EU RK-core suite 43.2 mm 45.2 mm (+5%) 34.3 mm (-21 %) • Operational COSMO-EU with LF shows small positive bias. • COSMO-EU with RK-core shows again a strong negative bias.

5. Statistical Scores vs 24h accumulation REGNIE March+April MJJA2009 • LF-core predicts too many 5-20 mm/24h events during winter. • RK-core has a dry bias and misses strong events during winter • During summer numer of strong events is overestimated when using the LF-core, but the RK-core has a dry bias. • ETS is similar during winter, but during summer RK gives lower ETS. •  Huge difference between both dynamical cores. ETS FBI

6. Dynamics and physics in COSMO-EU Motivation: • RK core leads to much less precipitation, especially stronger events are underestimated or missing completely. • Numerics experts tell us: ‘the vertical velocity in the LF-core simulations is much too noisy due to numerical problems, especially over orography’. • Model physics has over the last 2 decades been developed - and tuned - for the LF-core. Do we need to re-tune or completely overhaul our model physics?

7. Microphysics in COSMO-EU Possible modifications in cloud microphysics: • Higher fall speed of snow by changing der pre-factor a in vs = a (D/D0)b, (operational a=15, possible range a=15-25) • Higher autoconversion rate by reducing the number concentration of cloud droplets AU ~ Nc-2, (operational Nc = 500 cm-3, possible range 50-1000 cm-3) • Taking into account the density correction of the fall speeds of snow and rain v ~ (ρ0/ρ)1/2, (traditionally neglected in the COSMO model).

8. Precipitation accumulation for April 2009 RK control RK microphysics a=15, Nc=500 cm-3 a=25, Nc=200 cm-3 REGNIE • Only a very small effect for the total accumulation over one month.

9. accumulated precip in mm 24h accumulated precip 10. March 2009: NUMEX Exp. 6915 vs 6916 RK control RK microphysics a=15, Nc=500 cm-3 a=25, Nc=200 cm-3 REGNIE COSMO-EU Mean 7.1 mm Mean 9.9 mm Mean 7.2 mm Mean 7.3 mm Max 37 mm Max 33.4 mm Max 25.5 mm Max 36.8 mm • Orographic precipitation is enhanced by the changes in cloud microphysics • RK microphysics is the best forecast in this case

10. Statistical Scores 24h accumulation vs REGNIE for Feb + March 2009 FBI ETS • Strong overestimation of precip with LF-core, almost no bias in RK up to 5 mm/24h. • Re-tuning of microphysics can improve model behavior for strong precip events, especially orographic precip, but an underestimation of strong events still remains. • ETS is similar for LF-core and RK+Micro experiment.

11. accumulated precip in mm 24h accumulation 23. June 2009: (stand-alone simulations, no data assimilation) RK control RK microphysics a=15, Nc=500 cm-3 a=25, Nc=50 cm-3, (ρ0/ρ)1/2 REGNIE COSMO-EU Mean 4.3 mm Mean 4.8 mm Mean 4.0 mm Mean 4.0 mm Max 78 mm Max 102 mm Max 50 mm Max 55 mm • For this extreme event the underestimation by RK is a big issue for warnings! • Changes in microphysics have little impact on this kind of events.

12. Deutscher Wetterdienst GB Forschung und Entwicklung Conclusions • RK-core can solve the old problem of overestimation of precipitation in wintertime. • RK-core suffers from a dry bias. Strong and extreme events are underestimated. • A physically reasonable re-tuning of the microphysics parameterization can reduce some of the biases, i.e., lead to more orographic precip. But a lack/underestimation of strong events remains. • Re-tuning of convection scheme necessary? Maybe not. • Numerics experts will have to look into it again, but with a modified microphysics scheme the RK-core seems to give reasonable results. • The guidance for extreme precipitation events would change from over- to underforecasting. • What can we really expect for a model with 7 km grid-spacing?

13. Deutscher Wetterdienst GB Forschung und Entwicklung Additional slides ….

14. Accumulated precip May 2009: REGNIE LF COSMO-EU RK-core suite 74.5 mm 68.5 mm (-8%) 49.3 mm (-34 %) • Operational COSMO-EU with LF shows small negative bias. • COSMO-EU with RK-core reduces the precip amount even further, leading to a strong negative bias.

15. accumulated precip in mm 24h accumulated precip 25 March 2009: NUMEX Exp. 6915 vs 6916 RK control RK microphysics a=15, Nc=500 cm-3 a=25, Nc=200 cm-3 REGNIE COSMO-EU Mean 5.3 mm Mean 6.1 mm Mean 5.4 mm Mean 5.4 mm Max 34.4 mm Max 37.1 mm Max 23.2 mm Max 34.6 mm